Fix text_llm_runner kv cache pos count and use it for generate() (#15295)

### Summary
pos_ should advance by prefill and generated prompt size.
### Test plan
CI

cc @larryliu0820 @mergennachin @cccclai @helunwencser @jackzhxng

Co-authored-by: Hansong Zhang <[email protected]>

Author: pytorchbot

extension/llm/runner/text_llm_runner.cpp

@@ -183,11 +183,13 @@ Error TextLLMRunner::generate(
   // Generate max_new_tokens - 1 because prefill already generated 1 token.
   int64_t num_generated_tokens = ET_UNWRAP(text_token_generator_->generate(
       prompt_tokens,
-      num_prompt_tokens,
+      pos_,
       max_new_tokens - 1,
       temperature_ == -1.0f ? config.temperature : temperature_,
       wrapped_callback));
 
+  pos_ += num_generated_tokens;
+
   stats_->inference_end_ms = time_in_ms();
   if (!config.warming) {
     printf("\n");